HCV Death Rates 50% Underestimated
Limitations of conventionally derived chronic liver disease mortality rates: Results of a comprehensive assessment
Hepatology April 2008
M. Michele Manos 1 *, Wendy A. Leyden 1, Rosemary C. Murphy 1, Norah A. Terrault 2, Beth P. Bell 3
1Kaiser Permanente Division of Research, Oakland, CA
2Division of Gastroenterology, University of California San Francisco, San Francisco, CA
3Division of Viral Hepatitis, Centers for Disease Control and Prevention, Atlanta, GA
email: M. Michele Manos (email@example.com)
This work was presented in preliminary form at the 2004 American Association for the Study of Liver Diseases meeting.
Emerging Infections Program Cooperative Agreement with the U.S. Centers for Disease Control and Prevention
"By using a comprehensive list of mortality codes and considering HCC deaths as CLD related, we estimated the CLD mortality burden to be almost double that found using conventional death statistics methods. To fully assess the magnitude of CLD-related mortality, it is necessary to include death codes that will identify deaths attributed to chronic viral hepatitis, as well as those from hepatocellular carcinoma. Furthermore, our results suggest that the contributions of viral hepatitis and alcoholic liver disease to CLD mortality may be underestimated when based solely on death records, despite the use of a broad spectrum of CLD-related death codes."
Standard death certificate-based methods for ascertaining deaths due to chronic liver disease (CLD), such as the U.S. vital statistics approach, rely on a limited set of diagnostic codes to define CLD. These codes do not include viral hepatitis or consider hepatocellular carcinoma (HCC) deaths, and thus, underestimate the true burden of CLD mortality. Deaths associated with CLD may be further misunderstood because of the inherent limitations of death record information. Using a comprehensive list of CLD-related codes to search death records, we investigated the CLD mortality rate and associated etiologies (derived from medical records) in a large managed care health plan. From the 16,970 deaths among health plan members in 2000, we confirmed 355 (2.1%) as CLD related, including 75 with HCC. The age-adjusted CLD mortality rate using the comprehensive codes was 11.9 per 100,000 compared with 6.3 per 100,000 using only conventional codes. Based on medical records, the underlying CLD was attributed to alcoholic liver disease (ALD) in 44% of deaths, HCV infection with ALD in 16%, HCV without ALD in 18%, and chronic HBV infection in 7%. Only 64% of HCV-associated, 48% of HBV-associated, and 64% of ALD-associated deaths ascertained by medical record had that specific etiology mentioned on the death certificate.
Conclusion: CLD mortality burden was almost doubled by using a comprehensive list of mortality codes and considering HCC deaths as CLD related. Furthermore, the contributions of viral hepatitis and ALD to CLD mortality may be underestimated if based solely on death records.
In 2000, chronic liver disease (CLD) was reported by the National Center for Health Statistics to be the 12th most frequent cause of death in the United States and represented 1.1% of all reported deaths. For this and other conventional analyses, CLD-related deaths have been tabulated using International Classification of Disease codes (ICD-9 and eventually ICD-10) that focus on cirrhosis and alcoholic liver disease. Codes specific for viral hepatitis, an increasingly common cause of CLD, are not included in such mortality statistics. Deaths from hepatocellular carcinoma (HCC) are also not included despite indications that most deaths from HCC are attributable to CLD. Chronic hepatitis B and hepatitis C contribute to CLD-related deaths through both cirrhosis and as etiologic factors in HCC. The contribution of hepatitis C to CLD mortality is increasing, as is its contribution to HCC etiology. Although alcoholic liver disease remains an important contributor to CLD and associated mortality, it is likely to be increasingly present in combination with chronic viral hepatitis.
A recent U.S. study of CLD deaths during 1990-1998 utilized an expanded set of ICD-9 codes that included viral hepatitis and selected CLD sequelae to identify CLD-related deaths. Another study focused on hepatitis C-related deaths and supplemented multiple causes of death files with hospital discharge records. Both studies demonstrated that conventional methods substantially underestimate the number of CLD deaths and the proportion attributable to viral hepatitis. Those studies, however, did not include codes for HCC deaths or codes for additional chronic liver diseases such as fatty liver disease.
Regardless of the constellation of death codes chosen to define CLD-related deaths, assessments based solely on death records are inherently limited by the quality and completeness of those data. Thus, a more precise estimate of the contributions of viral hepatitis and other factors to CLD mortality may require investigation of both death records and medical records.
In this study, we compared 3 increasingly comprehensive lists of ICD-10 codes to identify CLD-associated mortality cases, and we used detailed health plan medical records to determine CLD etiology for each decedent. We report here CLD-associated mortality in 2000 among a diverse population of health plan members, describe the distribution of underlying CLD etiologies among those decedents, and examine the correlation between death record and medical record information.
This study was conducted within the Kaiser Permanente Medical Care Program of Northern California (KPMCP), currently providing health care to approximately 3.2 million members through 3,300 physicians, 35 outpatient clinics, and 19 hospitals. The program maintains extensive electronic databases including all outpatient pharmacy, inpatient, and outpatient encounter records (including diagnoses) and laboratory data for all tests performed (including pathology and radiology). The KPMCP membership is demographically similar to the U.S. Census-enumerated population in the Bay Area Metropolitan Statistical Area, except for the lack of representation of extremes in income. About 30% of the census-enumerated adults in the area are members of the health plan. The study was approved by the institutional review board of the Kaiser Foundation Research Institute, Oakland, California.
Identifying Potential CLD-Related Deaths in the KP Membership
Deaths of KPMCP members in 2000 (16,970 total deaths, crude mortality rate of 530 per 100,000 members) were identified from the annual California State Department of Vital Statistics single-cause-of-death mortality files. The all-causes-of-death data available in the National Center for Health Statistics Multiple-Cause Mortality files were then linked, and health plan members were identified through an established matching algorithm. Multiple-Cause Mortality Files included demographic information on each decedent, ICD Tenth Revision (ICD-10) codes for the underlying cause of death, and a list of up to 20 conditions contributing to the death listed on the death certificate. The year 2000 was chosen for study as it was midpoint in related ongoing population-based surveillance studies of chronic liver disease organized by the U.S. Centers for Disease Control and Prevention. Preliminary findings were reported previously in abstract form.
We defined potential chronic liver disease (CLD)-related deaths as those with an underlying cause-of-death code consistent with CLD. We used a list of codes that included conventional ICD-10 codes representing chronic liver disease and cirrhosis (corresponding to ICD-9 codes 571.0-571.9) and expanded ICD-10 codes representing viral hepatitis B and C and unspecified viral hepatitis (including codes for acute hepatitis B or C), plus selected sequelae of chronic liver disease such as hepatic coma, portal hypertension, and hepatorenal syndrome and a more comprehensive list of codes that included hepatocellular carcinoma, sequelae of viral hepatitis, toxic liver disease, liver failure, and other causes of CLD (Table 1). To identify probable hepatocellular carcinoma deaths, ICD-10 liver cancer codes C22.0 and C22.9 were used.
Medical and Demographic Data Collection
We collected age at death, sex, and membership status in 2000 from the KPMCP electronic patient databases and obtained race and ethnicity (Hispanic) information from the death certificate file. We classified all Hispanic patients as such, regardless of race. We collected clinical data from health plan electronic medical records for 1995-2000 including viral hepatitis and other laboratory test results, outpatient and inpatient diagnoses, and liver-related radiology and pathology reports.
Confirming CLD by Medical Record Review
Using all available electronic medical records, we determined whether each potential case had evidence of any type of CLD or hepatocellular carcinoma (HCC) within 5 years of death. Confirmation of CLD was defined as having at least 1 of the following: (1) 2 or more abnormal liver tests (alanine aminotransferase, aspartate aminotransferase, alkaline phosphatase, or bilirubin) at least 6 months apart within 3 years of death; (2) an inpatient or outpatient diagnosis within 18 months of death indicating CLD or HCC; (3) a radiologic report with evidence of CLD or HCC any time prior to death; or (4) a histology report with evidence of CLD or HCC at any time prior to death. We did not require that the CLD evidence from the medical record specifically match that described by the underlying cause of death, nor did we require substantiation that CLD or cirrhosis was the immediate cause of death. The goal was only to confirm that the case did have CLD or HCC in concert with a recorded CLD-related underlying cause of death.
Mortality was not considered associated with CLD if the decedent failed to meet any of the conditions listed above, or if a non-liver cancer metastasized to the liver, acute liver damage was evident, elevated liver tests were attributable to non-liver-related conditions, or polycystic liver disease was the only CLD evidenced. Cases with a sparse clinical record or who were rarely or never seen in the KPMCP system were excluded because of insufficient data (n = 26).
Cases with evidence of HCC in the medical record (histology, radiology, other clinical evidence) were defined as CLD-related deaths attributable to HCC, whereas all non-HCC deaths were considered to be a result of cirrhosis.
Assessing CLD Etiology
For confirmed CLD-related deaths, we assigned a definitive CLD etiology from the health plan electronic medical records for the period 1995-2000. Laboratory results and/or clinical diagnoses were used to assign etiologies. A hepatitis C etiology was assigned if a positive HCV antibody or HCV RNA test result or a clinical diagnosis of hepatitis C was found. Similarly, cases were attributed to chronic hepatitis B if either a positive hepatitis B surface antigen test or a diagnosis of chronic hepatitis B (in the absence of hepatitis B surface antigen testing) was found in the record (if done, a positive core antibody test was also required). Alcoholic liver disease was assigned as an etiology based on clinical diagnosis or treatment program participation. eOtherf types of CLD were evidenced by diagnoses and/or supporting laboratory results and included primary bilary cirrhosis, primary sclerosing cholangitis, cholestatic liver disease, sarcoidosis, hereditary hemochromatosis, autoimmune hepatitis, and fatty liver disease. Cryptogenic cirrhosis was assigned only when the medical record had sufficient evidence to exclude other known causes of CLD. Cases with multiple etiologies were assigned a single eprimaryf etiology for use in specific analyses (Fig. 1) based on this hierarchy: hepatitis C, hepatitis B, alcoholic liver disease, other CLD, unknown.
We also noted whether testing for viral hepatitis was evidenced in the electronic medical record. Such test results were crucial to rule out hepatitis B and C in cases with unknown or nonviral CLD etiologies. For chronic HBV infection we assessed hepatitis B surface antigen and total hepatitis B core antibody; that is, a negative hepatitis B core antibody test alone was accepted as ruling out chronic hepatitis B. For HCV infection, we assessed HCV antibody. Of note, a lack of records of viral testing could reflect that testing was never conducted, that the test was done outside the health plan at some point, or that testing was done in the health plan prior to 1995.
Death Record Notation of Etiology
We searched death record ICD-10 codes for reference to viral hepatitis (hepatitis C, hepatitis B, or unspecified viral hepatitis) or alcoholic liver disease and considered the code for underlying cause of death and all codes for additional causes of death. We concluded that death records reflected hepatitis C etiology if they contained a code for either hepatitis C or unspecified viral hepatitis, that they reflected hepatitis B etiology if they contained a code for either hepatitis B or unspecified viral hepatitis, and that they reflected alcoholic liver disease etiology if they contained any codes for alcohol-related disease.
We considered the health plan electronic medical record to be the most accurate source of information from which to assess the presence of CLD and to determine the underlying CLD etiology. If the death record identified the cause of death as CLD related, but the patient's medical record contradicted the presence of CLD, the case was excluded. Similarly, CLD etiologies were assigned based solely on medical record data.
We calculated age-specific CLD mortality rates among the KPMCP membership for 2000 by dividing the number of confirmed CLD deaths by the number of KPMCP members as of June 30, 2000. Rates were directly age-adjusted to the standard U.S. population in 2000, and 95% confidence intervals for those rates were calculated with the method based on the gamma distribution.
Death record accuracy was determined for cases with hepatitis B, hepatitis C, or alcoholic liver disease as one of their medical records-based CLD etiologies. For each etiology, if a mention of that etiology (as described) was included in any of the death certificate's cause of death codes, the death record was considered a match to the medical records-based etiology.
We used SAS (version 9.1, SAS Institute, Cary, NC) to identify and describe the sample of CLD deaths. We used the Pearson's chi-square test or the Mann-Whitney test to compare the characteristics of decedents with and without HCC.
Using a comprehensive list of ICD-10 codes likely to represent chronic liver disease (CLD)-related mortality (Table 1), we identified 428 presumed CLD-related deaths in 2000 among the Kaiser Permanente Medical Care Program of Northern California (KPMCP) adult membership. From a review of electronic medical records, we confirmed 355 (82.9%) as having CLD or hepatocellular carcinoma (HCC). Of the remaining decedents, 42 (9.8%) did not have evidence of CLD, and 31 (7.2%) had insufficient information. The 3 different methods of identifying potential CLD-related deaths (Table 1) provided different yields of confirmed cases. Of the 215 total deaths identified by the conventional codes, 191 (89%) were confirmed as CLD related, whereas of the 213 additional deaths identified by the comprehensive method, 164 (77%) were confirmed as CLD related. The lower specificity of the additional group of codes in the comprehensive method was a result in large part of deaths coded as unspecified liver cancer (C22.9), where only 19 of 32 decedents (59%) were confirmed as having had CLD or HCC.
Table 2 provides descriptive information for the decedents in total and also stratified by whether the death was HCC related or non-HCC related (presumably cirrhosis). Overall, cases were predominantly male (66%) and white, non-Hispanic (66%). The median age overall was 63 years (range 26-96 years) and did not differ substantially by sex. The median age of HCC-related deaths was significantly higher than that for cirrhosis deaths (68 years, range 38-96 years, versus 62 years, range 26-93 years; P = 0.04). A larger proportion of Asian cases was found in the HCC group versus in the cirrhosis group and represented primarily HBV-related cancer (11 of the 20 Asian HCC decedents). HCC deaths were more likely to be in men than were non-HCC-related deaths (P = 0.02). More than half the cases were identified by underlying cause-of-deaths codes for alcoholic liver disease (35%) or hepatitis C (16%). Codes for eacutef hepatitis C (B17.1) and hepatitis B (B16.9) contributed significantly to CLD death identification. In fact, the large majority (94%) of deaths with those underlying cause-of-death codes (8 of 9 with B16.9; 55 of 58 with B17.1) were confirmed as chronic liver disease associated. Liver cancer codes identified 75 confirmed CLD deaths, of which 68 (91%) were confirmed to be HCC and the other 7 as CLD but not HCC related.
The 355 CLD-related deaths represented 2.1% of the 16,970 total deaths in the membership in the year 2000. Of these 355 deaths, only 191 (54%) would have been identified using only conventional codes or 259 (73 %) using the expanded codes (Table 1; Fig. 1, column 1). Thus, the underlying cause-of-death codes unique to the comprehensive code list provided 96 (27%) of the cases.
CLD Death Rates
Table 3 presents the calculated CLD mortality rates for the KPMCP membership population, as well as those reported for the U.S. and California populations. Different KPMCP rates were calculated based on the numbers of deaths identified by each of the 3 increasingly comprehensive code lists to allow comparison of the methods within a single population. When based on conventional methods, the 2000 CLD age-adjusted mortality rate in KPMCP was less than that of the United States and substantially lower than that reported for the total California population, (6.3 per 100,000 versus 9.6 and 11.6 per 100,000, respectively). When KPMCP and U.S. rates based on an expanded code list were compared, KPMCP overall rates were also lower (8.5 versus 11.6). The KPMCP CLD mortality rate based on the comprehensive code list was considerably higher than that based on the expanded or conventional codes. Thus, for the KMPCP population, the comprehensive method revealed a mortality rate almost twice that found with conventional methods (11.6 versus 6.3). In the KPMCP comprehensive analysis, men had more than twice the rate of CLD-associated mortality as women, and differences between death ascertainment methods were reflected similarly in the rates for men and for women.
Underlying CLD etiology was deduced from health plan electronic medical records. Table 2 shows that many (61%) of the CLD deaths overall and most (73%) of the non-HCC deaths were of patients with alcohol-related disease. For many (45%) of the 121 hepatitis C-related deaths, alcoholic liver disease was identified in the medical record as an additional CLD etiology. Other CLD etiologies in the non-HCC deaths included 5 primary biliary cirrhosis, 5 autoimmune hepatitis, 1 cholestatic disease, 3 nonalcoholic fatty liver disease, 2 hereditary hemochromatosis, 1 primary sclerosing cholangitis, and 3 cryptogenic cirrhosis. The 2 other CLD etiologies among the HCC-related deaths were 1 sarcoidosis and 1 primary biliary cirrhosis.
The distributions of CLD etiologies for HCC and non-HCC cases are presented separately in Fig. 2. Contributions of hepatitis C with and without alcohol-related disease are shown. More than half the hepatitis C-related cirrhosis deaths were of patients with alcohol-related disease, whereas fewer than 20% of hepatitis C-related HCC deaths involved alcoholic liver disease. Only 5% of CLD deaths unrelated to HCC had an undetermined etiology, but one quarter of HCC deaths could not be assigned a CLD etiology, primarily because of limited diagnostic testing.
Of the total 27 CLD deaths in patients with known chronic HBV infection, more than half (56%) involved HCC. Of the 121 CLD deaths in patients with hepatitis C, 30% were HCC related. Only 4 of the 157 CLD deaths attributed to alcoholic liver disease without a known viral hepatitis contribution involved HCC.
In cases without a viral hepatitis etiology, we assessed whether HCV and chronic HBV infections had been ruled out by examining the health plan medical records for evidence of testing. Of the 210 cases without viral hepatitis etiology, 147 (70%) had a record of HCV testing, and 152 (72%) had a record of testing for chronic HBV.
Figure 1 also shows how the increasingly broad spectrum of death codes (Table 1) contributed to CLD deaths with specific underlying etiologies. For hepatitis C-related deaths, conventional methods would have identified only 21% of those found with the comprehensive code list. For hepatitis B-related deaths, fewer than 10% would have been identified conventionally. In contrast, most alcohol-related deaths (89%) would have been tabulated by the conventional approach. The additional codes provided by the expanded code list were important for identifying viral hepatitis cases, but the additional codes unique to the comprehensive list were necessary to identify more than 30% of hepatitis C- and more than 60% of hepatitis B-associated deaths. The comprehensive list also uniquely identified 21 of the 34 deaths with a defined CLD etiology other than viral hepatitis or alcoholic liver disease.
Death Record Capture of Etiology
On the death certificate, 64% of the hepatitis C-associated CLD deaths had hepatitis C or viral hepatitis noted, and 48% of the hepatitis B-associated deaths had hepatitis B or viral hepatitis mentioned. Similarly, 64% of alcohol-related deaths had alcohol-related liver disease noted on the death certificate. Age, race/ethnicity, and sex were not associated with death record versus medical record correlation regarding etiology (data not shown).
This study used a comprehensive list of International Classification of Disease death codes (ICD-10) to identify deaths associated with chronic liver disease (CLD) in members of a diverse, comprehensive managed care program. The findings were contrasted with those that would have been found using the limited, conventional CLD mortality codes that include only chronic liver disease and cirrhosis. Results were also compared with those that would have been found by a somewhat expanded list that included codes for viral hepatitis. Our comprehensive method of using a broad spectrum of death codes afforded the detection of almost twice as many confirmed CLD-related deaths as did the conventional approach. This suggests that methods used conventionally for U.S. national and state statistics substantially underestimate the true rates of CLD-associated mortality.
Additional CLD-related deaths found by the comprehensive method were identified primarily by death codes for hepatitis B, hepatitis C, and liver cancer. Other additional codes combined identified fewer than 7% of the total deaths. Interestingly, the ICD-10 codes for acute viral hepatitis were actually quite specific for chronic as opposed to acute liver disease. Reported previously, this apparently pervasive coding error contributes to faulty mortality statistics and needs to be corrected.
Death codes varied in their specificity for CLD (or HCC) deaths. For example, although the code for unspecified liver cancer (C22.9) identified 19 (5%) of the CLD-related deaths, it also identified 14 other deaths that were not confirmed as such (e.g., metastatic disease). The use of death codes with potentially lower specificity was not an issue in this study because medical records were available to confirm the presence of CLD or HCC. When confirmatory medical record resources are not available, the addition of the less specific underlying-cause-of-death codes may not be appropriate and could potentially lead to an overestimate of CLD-related mortality rates.
In our study population, more than 21% of the CLD deaths were identified by the inclusion of the selected liver cancer death codes. It is important to note that the relative contribution of liver cancer codes to subsequent CLD mortality estimates is dependent on the underlying hepatocellular carcinoma incidence and related death rates in the population being studied. The age-adjusted incidence rate of liver and intrahepatic bile duct cancer in the health plan in the year 2000 was 4.5 per 100,000, substantially lower than the 6.2 per 100,000 U.S. national estimate. Notably, liver cancer mortality rates vary greatly between states in the United States, ranging from 3.2 to 6.9 per 100,000 in 2000-2003.
Mortality cases without evidence of HCC were considered to be deaths due to complications of cirrhosis, based on each case having been assigned a CLD-related underlying cause of death on the death certificate. We did not seek to confirm cirrhosis from the medical record but only sought evidence that CLD was present. Possibly some of the non-HCC deaths actually did have undiagnosed HCC at the time of death. However, the dissimilarities between certain characteristics (for example, age, underlying CLD etiology) of HCC and non-HCC decedents suggest the groups are indeed distinct and that such unavoidable misclassifications may be minor.
Health plan death rates were compared with national and state rates as calculated based on conventional methods or with the expanded code list. The substantially lower KPMCP rate compared with national and state rates is perhaps not surprising. The health plan rates are derived from an insured population that does not include large representation from those with lower incomes. Furthermore, the KPMCP membership may not be representative of the Native American population in the health plan's geographic area. Although more than 5% of the San Francisco-Oakland metropolitan area is estimated to be Native American, less than 2% of the KPMCP membership is identified as such. Given that Native Americans have the highest CLD mortality rates of any racial-ethnic group nationally, their underrepresentation in the health plan may also contribute to our disparate rates. Regardless of how the KPMCP rates compare with those from larger populations, they demonstrate the importance of using a broader spectrum of death codes to accurately assess CLD mortality.
The death records often did not reflect CLD etiologies that were apparent from the medical record. Fewer than half the deaths that involved hepatitis B and fewer than two-thirds of those that were hepatitis C-related and alcoholic liver disease-related deaths mentioned that etiology in the death record. Thus, unless death record data are supplemented with other sources of medical history to assess CLD etiology, the contributions of viral hepatitis and alcoholic disease may continue to be substantially underestimated among CLD deaths despite the use of comprehensive death codes.
Our results also suggest that the lack of viral hepatitis testing may contribute to underestimation of related deaths. Of the cases not attributed to viral hepatitis, HCV and/or HBV infection had not been ruled out in more than a quarter. This observation suggests that up to an additional 17% of these 355 CLD-related deaths may actually be attributable to viral hepatitis. This limitation will remain despite the use of medical records to supplement death records and will improve only as viral hepatitis testing further penetrates the diagnostic assessment of CLD and HCC patients.
By using a comprehensive list of mortality codes and considering HCC deaths as CLD related, we estimated the CLD mortality burden to be almost double that found using conventional death statistics methods. To fully assess the magnitude of CLD-related mortality, it is necessary to include death codes that will identify deaths attributed to chronic viral hepatitis, as well as those from hepatocellular carcinoma. Furthermore, our results suggest that the contributions of viral hepatitis and alcoholic liver disease to CLD mortality may be underestimated when based solely on death records, despite the use of a broad spectrum of CLD-related death codes.